The effects of parameter settings on the performance of Genetic Algorithm through experimental design and statistical analysis

Abstract-Genetic algorithm (GA) is a meta-heuristic inspired by the efficiency of natural selection in biological evolution. It is one of the most widely used optimization procedure which has successfully been applied on a variety of complex combinatorial problems. The main drawback of GA, however, is its several tuning variables which need to be correctly set. The performance of GA largely depends on the proper selection of its parameters values; including crossover mechanism, probability of crossover, population size and mutation rate and selection percent. The objective of this research is to evaluate the effects of tuning parameters on the performance of genetic algorithm using the data collected as per Central Composite Design (CCD) matrix. To gather the required data, the proposed approach is implemented on a well-known travelling salesman problem with 48 cities. Then, regression modeling has been employed to relate GA variables settings to its performance characteristic. Analysis of Variance (ANOVA) results indicate that the function can properly represent the relationship between GA important variables and its performance measure (solution quality)